Local storage

The Go client keeps its offline source of truth in SQLite. On a standard managed installation, runtime files live below the XDG data home:

~/.local/share/devtrack/
├── data/
│   ├── db/devtrack.db
│   ├── learning/
│   ├── logs/
│   └── pids/
├── server/              # managed Python checkout and environment
├── devtrack.env         # generated runtime configuration
└── workspaces.yaml

XDG_DATA_HOME and the generated environment file can change these paths. Use devtrack settings and devtrack status rather than assuming a path.

Client and server storage are different

• Go client: local SQLite for triggers, work sessions, pending actions, history, MCP context, and offline replay.
• Python server: PostgreSQL is required for server persistence and server-side events.
• Voice retrieval: ChromaDB is optional and lives with the configured Python server (local in managed mode, server-side in external mode).
• Teams learning source: MongoDB is optional and used only when that integration is enabled.

The Go client never connects directly to PostgreSQL.

Outbound data

Pending-actions boundary

Actions that write to a PM platform, email, or Git remote are staged locally first with an explicit confidence score. They can be approved, edited, rejected, or allowed to expire according to the configured trust policy. DevTrack does not add a direct-send fallback.

Optional connections

• GitHub, GitLab, Azure DevOps, and Jira: requests contain the identifiers and content needed for the configured operation.
• Email, Telegram, Slack, and Teams: messages leave the machine only when the corresponding integration and action are enabled.
• External Python server: client-event synchronization is off by default and requires SERVER_EVENT_SYNC_ENABLED=true.
• Telemetry: off by default. Check with devtrack telemetry status; enable or disable it locally with devtrack telemetry on|off.

Credentials stay in the registered environment/configuration files and are never meant to be committed to source control.

AI processing

Local default

Ollama is the primary provider. When it is available, prompts and generated text are processed by the configured model at OLLAMA_HOST.

Optional cloud providers

OpenAI, Anthropic, and Groq are optional. They are used only when configured and selected by the provider chain. Prompt text required for generation is sent to the selected provider; do not enable a cloud provider if that is incompatible with your project policy.

Setup may offer an already-present OpenAI or Anthropic credential as a temporary fast lane while a local model downloads. It requires explicit acceptance, and Ollama remains primary.

Learning data

Git voice seeding reads local commit history. Additional PM, Teams, or recording sources are opt-in. Local profiles and examples are used to guide generated text and can be reset with devtrack learning-reset.

Your control

devtrack telemetry status
devtrack telemetry off
devtrack learning-status
devtrack revoke-consent
devtrack learning-reset
devtrack queue list
devtrack queue reject <id>
devtrack uninstall --keep-data

Before deleting data, stop the daemon and back up the path reported by devtrack settings. Database schemas evolve; direct SQL deletion is not a supported privacy workflow.

To report a privacy or security issue, open a GitHub issue without including credentials, tokens, private code, or personal data.

Security

• Use least-privilege, expiring tokens for integrations.
• Keep environment files and the XDG data directory readable only by your user.
• Use TLS and a strong DEVTRACK_API_KEY when the Python server is reachable over a network.
• Use a strong PostgreSQL password and do not expose PostgreSQL publicly.
• Review pending actions before extending auto-approval timeouts.
• Keep DevTrack, Ollama, PostgreSQL, and the host OS updated.

Last updated: September 3, 2026